He used GPT-3 to achieve “quantum speed reading”

This article is reprinted with the authorization of AI New Media Qubit (public account ID: QbitAI). Please contact the source for reprinting.

Have you ever flipped through a book until you were bald just to check a question?

If only if there were contents in the book, Just ask and you would know...Wait, "You will know if you ask", how can this happenChatGPTThat smell?

By the way, someone actually developed a Chat Bot based on this idea. The book used as an example is Marcus Aurelius's "Meditations" (the book has 12 chapters in total) Volume, 497).

In just a few minutes, AI can memorize the entire book.

Then just ask about something mentioned in the book, such as "How to deal with negative emotions?" AI can quickly scan the text and give high-quality answers, and even list them for you one by one.

As soon as the original post was posted, netizens flocked to like it.

Developer Dan Shippers said that it is not difficult to make this AI. The main contributor is the large language model GPT-3, plus a few more lines The code is done.

In addition, this buddy even smelled a business opportunity:

There are many copyright-protected text, audio and video data sets on the Internet, which can Made into a chatbot!

For people who want to obtain information, sometimes they may give up when they see that they have to sort it out slowly after being paid; but if AI can quickly lock the target content, there should be more people Be willing to pay out of pocket.

How to make it?

In Brother Dan’s words, the Bot production process is mainly divided into 3 steps.

First, of course, is to find the materials for the whole life. For example, the entire "Meditations".

Brother Dan downloaded the original "Meditations" from the Internet and saved it in a folder on Google Dive. Through Google Drive, users can store files in the cloud, synchronize and share them between devices.

Then run the following Python code on Google's cloud programming platform Colab. You can not only access Google Drive, but also easily interact with GPT-3.

Dan introduced that GPT-3 can create a chat bot for you based on any text material.

But there is a problem: the number of text words that can be input to GPT-3 at a time is limited.

So The second step is to access the folder stored on Google Dive, and then divide the entire text material into small pieces and save them into new files.

The third step, when the user asks a question, first access the small pieces of content in the book that contain relevant explanations through OpenAI’s API, and then add these content Pass it to GPT-3 and sort out answers in fluent language.

The specific code is as follows:

At this point, the chat robot specially customized for the entire "Meditations" is ready.

In addition, based on this basic idea, Brother Dan also made a certain series of podcasts he liked into a chat bot.

He said that he once heard an interesting term on the show, but forgot what it meant. Usually if I want to understand it, I have to listen to an episode again.

But now there is a chatbot - hard of hearing? nonexistent.

Who is Dan?

Having said that, Dan’s past experience may also have had a certain impact on people being able to see new business opportunities through the chat machine.

He graduated from the liberal arts philosophy department of the University of Pennsylvania, but this guy is actually a technology tycoon.

He started programming in the 5th grade of elementary school. During college, he and his friends founded a software company, FireFly, with an income of more than six figures.

When he was still in his sophomore year, many technology companies had already offered him an olive branch, but they all rejected him, saying they wanted to complete his studies first.

As soon as he graduated from college, he sold FireFly to Pegasystems, a well-known business software company, and made his first pot of gold.

Immediately afterwards, he served as the project leader at Pega and increased the revenue of the business unit to the million-dollar level.

Later, he left Pega and founded a company Every in 2020, which mainly creates business briefings and podcasts. It currently has 60,000 subscribers.

(No wonder he would think of turning the podcast content into a chatbot

)

In addition to starting his own company, Dan also serves as the head of Sequoia Capital. Scout, advising venture capitalists.

The above is the detailed content of He used GPT-3 to achieve “quantum speed reading”. For more information, please follow other related articles on the PHP Chinese website!